Multidimensional Analytical Study of Heart Sounds: A Review

INT. J. BIOAUTOMATION, 2015, 19(3), 351-376

351

Multidimensional Analytical Study

of Heart Sounds: A Review

Akash Kumar Bhoi

1*, Karma Sonam Sherpa

2, Bidita Khandelwal

3

1Department of Applied Electronics and Instrumentation Engineering

Sikkim Manipal Institute of Technology

Majhitar, India

E-mail: [email protected]

2Department of Electrical and Electronics Engineering

Sikkim Manipal Institute of Technology

Majhitar, India


3Department Gen Medicine

Central Referral Hospital and SMIMS, SMU

Gangtok, India E-mail: [email protected] *Corresponding author

Received: April 15, 2015 Accepted: September 23, 2015

Published: September 30, 2015

Abstract: Heart diagnosis by phonocardiography and auscultation is highly dependent on

experience and there is a considerable inter-observer variation. The complex structure of the

Phonocardiogram (PCG) and the variations due to cardiac contractility can generate

additional difficulties for auscultation. This review paper focuses on such critical problem

solving issues with a variant of analysis. However, different methods and techniques are also

described for detection and analysis of PCG signal and it will certainly aid findings in novel

computational tools in biosignal processing.

Keywords: Phonocardiography, Variant Analysis, Multidimensional Analysis.

Introduction Heart sounds and murmurs arise as a consequence of turbulent blood flow and vibrating

cardiovascular structures. Some decades ago, the Phonocardiography (PCG) has been largely

used together with the Electrocardiography as providing important information about the

cardiac diseases. Later, the ultrasound techniques took its place offering a lot of additional

quantitative indices related to the cardiovascular system. Recently the interest to the PCG and

other more sophisticated heart sounds investigations reappears in the field of medicine [88].

The reason is that they are non-invasive, economical and accurate methods for assessing

different heart valve pathologies. Hult et al. [88] suggested that the detection of a third sound

in adults is shown as simple method for systolic heart failure discovery.

The PCG signal discloses information about cardiac function through vibrations caused by the

working heart. In the early days of PCG signal analysis, manual interpretation of waveform

patterns was performed in the time domain. Heart sounds were identified as composite

oscillations related to valve closure and heart murmurs seemed to derive from malfunctioning

valves or from abnormal holes in the septal wall [1]. Heart auscultation, the technique of

listening to heart sounds, is a convenient and economical method for diagnosing

cardiovascular diseases (CVD). Though clinical diagnosis of cardiovascular disease mainly

mailto:[email protected]




352

refers to other methods such as electrocardiography (ECG), angiocardiography and so on, the

considerable role of PCG in diagnosis of CVD cannot be underestimated, especially in

computer-aided CVD diagnosis of fusing other vital signals (e.g. ECG) to improve the

diagnostic accuracy [66]. The physiological variability of the mechanical function of the heart

is reflected in the produced acoustic vibrations – the heart sounds. Heart sounds have been

widely used in clinical practice since the introduction of the first stethoscope by Laennec in

1816, and the invention of phonocardiography, the graphic recording of heart sounds, by

Einthoven in 1894. Heart sounds and their clinical utilization in cardiovascular and

cardiopulmonary diseases have been extensively studied for many years [36]. Relations

between morphological features of heart sounds and hemodynamic parameters have been

quantitatively described in both animal models and humans [37, 38].

Heart sounds and murmurs are of relatively low intensity and are band-limited to about

10-1000 Hz, (Fig. 1).

Fig. 1 Relationship between the acoustic range of cardiac sounds and the threshold

of audibility of sound pressure for human ear

(figure redrawn from Leatham [2])

The traditional areas of auscultation (Fig. 2), where the radiated sound intensity from each of

the four heart valves is maximized, are defined as [5]:

Mitral area: The cardiac apex.

Tricuspid area: The fourth and fifth intercostal space along the left sterna border.

Aortic area: The second intercostal space along the right sternal border.

Pulmonic area: The second intercostal space along the left sternal border.

Fig. 2 The traditional auscultatory areas on the chest:

M (refers to the mitral area), T (refers to the tricuspid area),

P (refers to the pulmonic area), and A (refers to the aortic area).


353

Fig. 3 Wiggers diagram, showing pressures and flows in the left side of the heart

over one heart cycle and their relation to electrical (ECG) and mechanical (PCG) activity

The first heart sound (S1) and second heart sound (S2) in Wiggers diagram (Fig. 3) are easily

audible in nature, and the time duration is around 150 ms and 120 ms. It ranges from

20 to 150 Hz. Heart sound (S1) is associated with the closure of the mitral-tricuspid valve, it

occurs during the isovolumetric contraction of the ventricles. Heart sound (S2) is related to

the aorticpulmonary valve at the time of the isovolumetric relaxation of the ventricles.

The third heart sound (S3) and fourth heart sound (S4) are very light sound, i.e., almost

inaudible in nature. S3 and S4 are very low frequency sound. S3 is not originated from the

valve although it happens at the beginning of the diastole and hence, it is known as Proto-

diastolic sound. Bulk of blood flow into the left ventricles causes vibrations in the valve.

Heart sound (S4) is mainly found in the healthy children and not, usually found in adults. It is

called pre-systolic gallop in pathological term when found in adult. The Healthy signals are

the signals which have a clear sound of S1 (‘lup’) and S2 (‘dup’) and almost negligible sound

of S3 and S4. Healthy heart sound contains heart sound segment S1 and S2, which gives

information of functionality of heart sound [3, 4]. The intensity of the first heart sound (S1)

varies with certain conditions. Those who suffer from emphysema, obesity, pericardial

effusion, myocardial disease or mitral regurgitation may have decreased first heart sound

(S1). Increase in intensity of the first sound depends upon the vigour of the ventricular

systole. Thus results in loud sounds in exercise, emotional states and in hyperthyroidism,

anemia, mitral stenosis and hypertension. The conditions giving rise to change in intensity of

the first heart sound (S1) has been discussed by various authors [89-92].

The relationship between ECG and PCG can be figured out from Wingger diagram (Fig. 3)

where, S1 occurs with low frequency vibrations approximately 0.05 second after the onset of

QRS-complex of ECG signal. S2 starts approximately 0.03-0.05 second after the end on

T wave of the ECG. S3 starts at 0.12-0.18 second after the onset of second heart sound (S2)


354

and the fourth heart sound (S4) starts approximately 0.12-0.18 s after the onset of P wave of

ECG signal. S1 and S2 have two major components, M1 and T1, A2 and P2, respectively.

M1 is caused by mitral closure and blood flowing interruption in left atrial and systolic

ventricular and T1 is caused by tricuspid closure. A2 starts before aortic valve closed, and

P2 starts after pulmonary value closure, both are caused by intraventricular pressure dropping

and blood returning in diastole. The delay time between M1 and T1, A2 and P2 are called the

first split and the second split, respectively. The measurement of the first and second split,

lower or higher than 30 ms, will easily make it possible to make discrimination between the

normal or pathological type [68]. It is often followed by echocardiography during the

abnormal auscultatory findings. However, the lack of reliability of ordinary auscultation its

expense and awkwardness of echocardiography make it desirable to develop a more practical,

inexpensive, reliable and non-invasive approach to auscultation, one that could also be

adapted for continuous monitoring [71-75]. Akbari et al. [76] performed Digital Subtraction

Analysis of the heart murmurs signal using a custom computer program called Murmurgram

for the detection and characterization. In essence, this program subtracts the recorded sound

from two adjacent cardiac cycles to produce a difference signal, herein called a

“murmurgram” [76].

Feature extraction is typically a preceding step for a classification or regression task.

Heart sound classification, based on morphological spectral and time–frequency features, has

been previously used for assessing the condition of bioprosthetic heart valves [40-42].

Auscultation, the noninvasive cardiac testing, is used as a primary detection tool for diagnosis

of heart valve disorders since invention of stethoscope in 1816 by Lannec [47]. In Bender,

it is reported that few heart valve diseases are best detected only by means of auscultation

process [48]. Auscultation is the most common and cost-effective technique, continues to

provide an important source of clinical information related to heart valves and also, cannot be

totally replaced by alternative technical methods like echocardiography [49]. In case of

abnormal heart sounds, there could be several other sounds in the PCG signal besides primary

heart sounds. Murmurs are abnormal heart sounds and refer to different pathological

conditions as per location, shape, duration and other associated features [52]. Murmurs are

generally high-frequency, noise like sounds that are produced as a result of turbulent blood

flow. Different features of PCG signals like intensity, frequency content, split information,

time relations etc. are helpful in detecting heart valve diseases, if any and the state of the heart

function [53]. Ian Cather has presented Artificial Neural Network (ANN) as a discriminative

model for classification of five different heart sounds taken from 48 recordings of nine

different subjects using wavelet based feature extraction technique [54]. Ölmez et al. [55]

have given a classification technique that utilizes Daubechies-2 wavelet detail coefficients at

the second decomposition level for classification of seven different heart sounds collected

from 28 subjects using ANN. Reed et al. [56] have described a computer-aided diagnosis

mechanism for five different pathological cases using seven level wavelet decomposition,

based on a Coifman fourth order wavelet kernel and Ari et al. [57] proposed, a binary decision

on heart sound, whether pathological or not, in a Digital Signal Processor based system. Choi

[58] proposed a technique for detection of valvular heart sounds as normal or pathological

using wavelet packet decomposition and support vector machine with fifth order polynomial

kernel function. Information, such as the temporal localization of the heart sounds, the

number of their internal components, their frequency content, and the significance of diastolic

and systolic murmurs, could all be studied directly on the PCG signal. In order to recognize

and classify cardiovascular pathologies, advanced methods and techniques of signal

processing and artificial intelligence need to be used. The advancement of technology has

paved the way for signal processing methods to be implemented and applied in many simple


355

tools useful in everyday life. This is most notable in the medical technology field where

contributions involving the intelligent applications have boosted the quality of diagnosis.

Proposing an objective signal processing method capable to extract relevant information from

biosignals is a great challenge in telemedicine and auto-diagnosis fields [86].

Sa-Ngasoongsong et al. [87] presented the design and testing of a wireless sensor system

developed using a Microchip PICDEM developer kit to acquire and monitor human heart

sounds for phonocardiography applications. This system can serve as a cost-effective option

to the recent developments in wireless phonocardiography sensors that have primarily focused

on Bluetooth technology. This wireless sensor system has been designed and developed in-

house using off-the-shelf components and open source software for remote and mobile

applications. The small form factor (3.75 cm × 5 cm × 1 cm), high throughput (6,000 Hz data

streaming rate) and low cost ($13 per unit for a 1,000 unit batch) of this wireless sensor

system make it particularly attractive for phonocardiography and other sensing applications.

The experimental results of sensor signal analysis using several signal characterization

techniques suggest that this wireless sensor system can capture both fundamental heart sounds

(S1 and S2) and is, also, capable of capturing abnormal heart sounds (S3 and S4) and heart

murmurs without aliasing. The results of a denoising application using Wavelet Transform

show that the undesirable noises of sensor signals in the surrounding environment can be

reduced dramatically. The exercising experiment results also show that this proposed wireless

PCG system can capture heart sounds over different heart conditions simulated by varying

heart rates of six subjects over a range of 60-180 Hz through exercise testing [87]. Shub [50]

compared the cardiac physical examination with echocardiography for evaluating systolic

murmurs and concluded that echocardiography is not required for all patients with systolic

murmurs and should not replace cardiac physical examination; Chizner [51] reviewed the

fundamental principles of the art of cardiac auscultation and emphasized on the proper use of

the stethoscope and the diagnostic and prognostic significance of the myriad heart sounds and

murmurs present in patients with and without symptomatic heart disease, and Debbal et al.

[67] proposed computerized analysis of heart sounds which is concerned with a synthesis

study of the Fast Fourier Transform (FFT), the Short-time Fourier Transform (STFT), the

Wigner distribution (WD) and the Wavelet Transform (WT) in analyzing the

Phonocardiogram signal (PCG). It is shown that these transforms provide enough features of

the PCG signals that will help clinicians to obtain qualitative and quantitative measurements

of the Time-frequency (TF) PCG signal characteristics and consequently aid diagnosis.

Physiological importance of heart sounds

Fetal maturity As during pregnancy all the frequency determining parameters undergo physiological change.

The growing dimensions and increasing contractile strength of the myocardium characteristic

changes of the PCG spectrum are to be expected as a function of the stage of fetal maturity.

In order to verify this assumption, extensive experimental studies have been done. The power

spectra were integrated in order to compare different spectra using only one single parameter.

The determining measure was the frequency within which 80% of the total PCG power is

contained, i.e., where the power spectrum integral reaches 80% of its maximum. Mean values

of the 80% marks for the first and second heart sound are shown in Fig. 4. Each point was

computed from 30-50 fetuses of the same gestational age. Of course, mean values have been

computed only from spectra of fetuses that were known to be developing normally.


356

Fig. 4 Relationship between fetal PCG spectrum and gestational age.

The traces show the mean 80% power marks of the integrated power spectra

for the first and second heart sound dependent on the week of pregnancy.

The standard deviation of the single measuring points typically corresponds to 1.4 weeks.

(figure redrawn from Nagel [16])

The first heart sound reveals a steady shift towards higher frequencies up to about the 34th

week of gestation. Thereafter, the shift reverses its direction. The spectrum of the second heart

sound steadily shifts to lower frequencies. The decrease in frequencies can be explained by

the growth of the heart from physical models, whereas the ascending slope of the first heart

sound's spectrum is primarily caused by the predominance of the increasing contractile

strength of the myocardium during that stage of pregnancy. On the basis of these findings,

spectral analysis of the PCG can be considered a real alternative to better known

measurements such as ultrasonic imaging in the determination of fetal maturity.

When investigating the reliability of this measurement, possible disturbing influences such as

changing cardiovascular conditions (e.g. the heart rate) have to be considered. So far,

however, no negative effects have been found. Indeed, measurements were always done at the

basal heart rate with no labor activity, and in the absence of accelerations and decelerations.

In several cases the spectra did not fit any stage of maturity, showing distinct deviations from

normal values. Further investigations (done partially after birth), confirmed the existence of

heart diseases, predominantly valve malfunctions. This diagnostic capability is the direct

consequence of the fact that each pathological variation of either cardiac function or anatomy

results in a change in the PCG spectrum.

Cardiac contractility variability Variation in the amplitude of the first heart sound is seen in almost every sample of the

Phonocardiogram (PCG). During heart-sound signal processing, Xiao et al. [29] found that

this phenomenon had its own regularity. There are many factors causing this variation, such

as respiration, exercise, psychological activity, drugs, temperature, smoking, disease, etc.

These factors can affect the heart's state of inotropism, chronotropism, and dromotropism,

which will be reflected in the PCG. It has been reported that changes in the amplitude of the

first heart sound are closely correlated with the maximum rate of rise of left ventricular

pressure (a standard measure of cardiac contractility) [23-28].

Descriptive analysis have already being done based on the relationship between the amplitude

of the first heart sound (S1) and the cardiac contractility [23-28], Xiao et al. [29] proposed a

concept of Cardiac Contractility Variability (CCV) and a method of analysis. This study was

carried out to examine the regularity of the amplitude variability of the first heart sound and to

evaluate the practical significance of the CCV. Cardiac contractility variability can easily be


357

studied by means of heart-sound analysis, a safe, non-invasive, and inexpensive technique.

Analyzing the regularity of S1 variability and evaluating CCV may have practical utility and

is worth further study. A study of CCV may be of help in gaining insight into the neurogenic

cardiovascular state of patients in different clinical conditions.

A change in the pattern of CCV may be an indicator of systolic function. Also, it might be an

indicator of lung ventilation function e.g., in a patient with emphysema, the compliance of

lungs and thorax decreases. In turn, the respiratory variation of venous return to the heart is

reduced, and therefore the variation of S1 amplitude in a respiratory cycle becomes less

obvious.

Spectral analysis of prosthetic heart valve sounds Analog techniques for processing prosthetic valve Phonocardiograms (PCG) have met with

limited success in extracting this information, because of their poor spectral resolution and

lack of versatility. Numerical methods of signal processing overcome most of these

limitations, but the need for a computer to implement numerical methods raises the question

of cost-effectiveness in many applications. Numerical analysis of prosthetic valve signals has,

therefore, received very little attention outside the academic and laboratory context.

Cost reductions in computer hardware arising from the use of micro-processors; make it

possible to envisage dedicated clinical instruments for processing prosthetic valve sounds in

view of assessing overall valve performance and detecting component degradation at an early

stage. Basic spectral considerations for the design of such instruments are discussed [15].

From the very beginning of phonocardiography, it has been known that sound-producing

events which give rise to the PCG include oscillations of blood masses, movements of the

heart wall and valves and turbulence in blood flow. Signals recorded from the body surface

are further conditioned by the transmission characteristics of intervening tissues and the

transducer used. Because these structure (both the active sound producing sources and passive

components) differ morphologically. Their vibrational modes of resonance are also different.

Fortunately, most prosthetic valve sounds are much louder and contain higher frequency

components than the other structures of the heart so that extracting them from the PCG is

relatively easy.

In the past, two techniques of spectral analysis (spectroanalysis) have been used for detecting

such changes. One technique, called “octave-band analysis” uses a bank of band-pass filters

and an electronic system that evaluates the relative energy of the signal output by each filter

[6]. The term “octave-band analysis” arises from the fact that the central frequencies of two

successive filters are in a ratio of 1:2. An average curve is then obtained by plotting the

relative amplitude of prosthetic sounds versus frequency over many cardiac cycles. The other

technique named “sound spectrography” was developped by Bell Laboratories for speech

analysis; it was first used in phonocardiography by Geckeler et al. [7] and McKusic et al. [8]

in 1954. Instruments based on this technique provide a three dimensional display of

frequency, amplitude and time for a short segment of PCG recorded on a loop of magnetic

tape to reproduce the PCG with the periodicity required for the analysis. The principle of

analysis is to use a single fixed band-pass filter and to translate the spectrum of the signal in

successive steps. This is achieved through amplitude modulation of the PCG by a sine wave

whose frequency is increased step-wise. The spectrum of the PCG is thus progressively

swept; the amplitude of each frequency band is evaluated and recorded on photographic paper

as dots of variable intensity. A drawback of this instrument was that only a qualitative

evaluation of spectral energy was possible from the gray-scale graph. This was overcome by a

later instrument introduced by Winer (1965) which presents the signal information as


358

frequency-time contours of equal spectral energy. It is this later technique that has been

mostly used in the non-invasive evaluation of prosthetic ball-valve variance, as reported by

Hylen et al. [9] (1969), Aigner et al. [10, 11] (1973-1977), and Kagawa et al. [12] (1977).

An example of the spectrum of an aortic valve sound obtained by this technique is presented

in Fig. 5.

Fig. 5 Contour spectrogram of the PCG of a Harken aortic ball valve:

Aortic opening sounds (AOS), Aortic closing sound (ACS).

(figure redrawn from Durand et al. [15])

Despite the fact that such a representation contains a lot of information on the prosthetic valve

being evaluated; the extraction of quantitative parameters is quite difficult. The spectral

analysis of prosthetic valve sounds done by octave-band analysis and contour spectrography

is mostly qualitative and generates quantitatively the same parameter: the highest harmonic

component of opening (OS) or closing valve sounds (CS). Therefore, the information

contained in the Phonocardiogram of cardiac valve prostheses. In addition, being essentially

analog techniques and the difficulty of implementing filters with near ideal band-pass

frequency response, it suffers from limited frequency resolution. While these techniques were

finding some limited clinical acceptance whereas, the increasing availability of digital

computers in the early '60s promoted the development of numerical methods of signal

analysis. One of the most important events of this decade was the introduction of the FFT

algorithm by Cooley and Tukey [13] in 1965, which, by drastically reducing the computing

time required for spectral evaluation, practically revolutionized signal analysis.

An efficient FFT implementation of the Discrete Fourier Transform (DFT) is crucially

important since the DFT is the basis for most spectrum analysis techniques: a speed

improvement of roughly 100 in the computation of a 1024 sample sequence can result from

the evaluation of the FFT versus the evaluation of spectral components by a straightforward

numerical integration. The Discrete Fourier Transform pair can be represented by:

1

0

, 0, 1, , 1N

nk

N

n

X k x n W k N

(1)

1

0

1, 0, 1, , 1

Nnk

N

n

X n x k W k NN

(2)

where the amplitude and phase information of the harmonic term 2 /j N

nW e is contained in

the complex coefficient X(k).


359

In order to retain some temporal information of the PCG signal, the concept of the short-time

Fourier representation of a signal is quite useful [14]. Mathematically, it can be defined as:

1 2

0

, , for 0,1, , 1 and 2

N j mk

N

m

NX n k W m x n m e k N n

(3)

where W(m) is a real window sequence of duration N centered on the time index n in order to

determine the portion of the input signal that receives emphasis at a particular value of n.

The procedure is illustrated in Fig. 6.

Fig. 6 Schematic illustration of the spectrogram construction of a signal x(n)

(figure redrawn from Durand et al. [15])

The short-time DFT is clearly a function of two variables: for each value of the time index n,

the obtained Fourier representation (index k) of its neighboring values emphasized by the

window sequence used. Thus, by taking successive short-time DFT at each or multiple values

of time index n, it is possible to construct a digital time-dependent frequency representation of

the input signal.

PCG spectral composition The heart represents a rather complex oscillatory system, whose spectral power distribution is

determined by numerous factors, such as the dimensions of the vibrating tissues (muscle and

valves), their properties with respect to elasticity and density, and the physical conditions of

vibrational excitation, especially the spectrum of the excitational function and the tension of

the tissues that is caused by external forces as well as by their own contraction. Additionally,

the spectrum is influenced by the surrounding media, primarily the blood contained in heart

and vessels. Last but not least, the intracardial blood volume itself represents a vibrational

system, whose contribution to the heart sounds appears particularly distinct in the case of

eddy flows caused by cardiac malfunctions, such as valvular stenosis or insufficiency. Due to

the complexity of the heart it is not possible to formulate an analytical description of the


360

vibrations or their spectra. The well-known Laplace model describes the ventricles as

a thin-walled hollow sphere. The tension (s) of the wall depends on the trans-mural filling

pressure (p), the radius (r) of the sphere, and the thickness (h) of the wall, that is the

myocardium:

2

prs

h (4)

For thick-walled structures, the relationship is

33

3

3 3

12

( )

ai i

a i

rp r

rs r

r r

(5)

where ri and ra represent inner and outer radii of the wall, and r lies between these limits.

The tension decreases from the inside towards the outside. In the course of the heart cycle,

both wall tension and intracardial pressure and also the radius of the ventricles are subject to

change. According to the stage of heart action, these three variables exchange the function of

dependent and independent parameters. During diastole there is a passive increase of wall

tension (the preload) caused by the filling pressure, whereas during the systole the contraction

of the myocardium actively increases the wall tension (the after load) and thus produces an

interventricular rise of pressure. During the ejection time, the tension of the myocardial wall

is influenced essentially by the decrease in ventricular radius and the increase in its thickness.

Besides pathologic variations of the PCG power spectrum, characteristic shifts are also to be

expected according to physiological changes of the characteristic parameters that occur

depending on gestational age and fetal maturity [16].

Representation of heart sound signal in time-frequency domains Several signal models can be found in literature for the decomposition of heart sound signal,

such as the chirp models [60, 61], the damped sinusoidal models [62, 63], the modified Prony

models [64], and Leung et al. [65] employed the Gaussian modulation model to decompose

the second heart sound for the diagnosis of pediatric heart diseases. For each patient, the splits

of 20 successive cardiac cycles are measured; their mean and standard deviation are then

calculated and used to characterize the two splitting patterns. It is found that the two simple

statistical quantities can be used to identify the splitting patterns and hence, offer important

diagnostic information.

2 2( ) / 2

1

cos(2 )m

mi mi

Lt t

m mi mi mi

i

h t a e t

(6)

where hm(t) is the heart sound signal of the mth

cycle. Namely, Eq. (6) means that hm(t) is the

sum of Lm atoms. Every atom is characterized by five parameters: tmi is the time delay of the

ith

atom with respect to the start of the mth

cycle; ami is the amplitude; ωmi is the frequency;

σmi is the time width that the atom needs support; βmi is the phase. Therefore, the heart sound

signal of this cycle is represented by the set of atoms {tmi, ami, ωmi, σmi, βmi, 1 ≤ i ≤ Lm}.

The number of atoms, Lm, and the five parameters for each atom can be obtained using short-

time Fourier transform (STFT) analysis, as described in [65].


361

The STFT of the heart sound signal hm(t) is

2, mH t f h t t e d (7)

Biometric identification The performance of traditional biometric identification systems is, as yet, unsatisfactory in

certain applications. For this reason, other physiological or behavioral characteristics have

recently been considered, using new electrical or physical signals linked to a person’s vital

signs. Francesco Beritelli examines the biometric characteristics of PCG signals from cardiac

auscultation. The idea is that PCG signals have specific individual characteristics that can be

taken into consideration as a physiological sign used in a biometric system [31]. The database

used for the biometric identification by Francesco Beritelli, contains heart sounds from people

suffering from various types of cardiac pathology. In order to study the spectral characteristics

of heart sounds and, therefore, their biometric properties, the recordings which are used

contains heart sounds relating to the following pathologies:

1) innocent systolic murmur; 2) mitral regurgitation variations; 3) mitral regurgitation;

4) mitral stenosis; and 5) third heart sound.

Biometric technologies are based on the use of individual characteristics for the recognition or

identification of an individual [30]. They are divided into two areas:

1) Physiological characteristics (unique and unvarying), which include the geometry

of the hand and the palm print, fingerprints, retina, or iris image, and the (geometrical)

features of the face;

2) Behavioral characteristics (unique but varying), which include signature, way

of walking, voice (the latter also belongs to the previous group), and keyboard typing style.

A feature common to all biometric technologies is the capability of human recognition from

biometric data. This consists of a series of basic processes:

i. Acquisition and storage of reference biometric data acquired by means of sensors

(optical, ultrasonic, thermal, etc.);

ii. Acquisition of new biometric data at the start of a recognition process, for

comparison with the reference data;

iii. Determination of the correspondence of the newly acquired data to the stored

reference to determine whether they both could have been generated by the same

person.

In the first phase (enrollment), a sample is acquired, which might be converted to a template,

a model, or left unprocessed. This is accomplished in a controlled environment so as to

guarantee the security of the original and the properties of the biometric print. The second

phase (matching) is activated whenever it is necessary to verify or identify a print.

Authentication or verification is performed by comparing the acquired print with a reference.

Identification is performed by searching in an archive for a compatible template.

They therefore serve to identify a person whose identity is not necessarily known a priori.

The implementation of a biometric solution requires recognition threshold values to be

assigned. Due to various physical and environmental factors, in fact, the result of a biometric

identification process is never completely certain. The definition of the recognition threshold

establishes the boundary between acceptance and rejection of a request. In this way, it is

possible to set the security level on which the False Reject Rate (FRR) depends on (i.e., the

number of times the system does not recognize a sample as coming from the same individual

who produced the reference) and the False Accept Rate (FAR) (i.e., the number of times the


362

system incorrectly matches a sample from one person to a reference from another).

The FRR and FAR are generally considered as indices of the biometric system’s performance.

The segmentation algorithm, which is essential for the subsequent matching phase, generally

segments the sequences analyzed correctly and so is very robust to the degradation typically

occurring in PCG sequences recorded in a real application context. As far as the matching

phase is concerned, first conducted a preliminary study to identify the heart auscultation

region (i.e., one of the four auscultation positions for the stethoscope), which gives the best

degree of separability between classes for intra and interperson distances.

Matching algorithm

The metric used to measure the distance between the two signal spectra was Euclidean.

Considering the spectra as N-dimensional vectors

2

1

1, ( )

N

i i

i

d X Y X YN

(8)

where X and Y are the vectors containing samples of the signal spectrum whose distance is to

be measured, and is the number of samples calculated for each frame. When S1 and S2

signals are extracted from a cardiac sound recorded from the same individual, the distances

with respect to the spectra are expected to yield lower values than those obtained when S1 and

S2 spectra extracted from sounds recorded from different individuals are used. Following a

series of recordings, the distributions of the distance variable with S1 recordings made from

both the same person and different individuals is determined.

Quantitative methods for PCG signal analysis

Matching pursuit method

Zhang et al. [18] developed a time-frequency scaling transformation based on the Matching

Pursuit (MP) method for the Phonocardiogram (PCG). The MP method decomposes a signal

into a series of time-frequency atoms by using an iterative process [18].

A. Time-scaling of PCG signals

The MP method represents a signal as a combination of an infinite number of time-frequency

atoms [17]. It can be written as

0

i i

i

x t a h t

(9)

with

i i i ih t g t u t (10)

and

cos 2

ii

i

i i

t pg t g

s

u t f t

(11)


363

where ai are the expansion coefficients. The parameters (si) (the scale factors) are used to

control the width of the waveform envelope, and (pi) are used to specify their temporal

location. The parameters βi are normalizing factors to keep the norm of hi(t) equal to one.

The purpose of time scaling the PCG is to change the rate of presentation while keeping the

perceptual quality of the original signal. For a uniform change in the time scale, the time t of

the original PCG is mapped into the transformed time scale t through the mapping

t t (12)

Zhang et al. [18] found that the time scale expansion is more useful than time scale

compression. Thus, γ is always larger than one. In the MP method, the temporal properties of

a time-frequency atom are related with the time-position (pi) and the scale (si) which are

modified for time-scaling of the PCG. For an input signal x(t), the reconstructed signal x′(t) by

the MP method for m time-frequency atoms is given by

1

0

cos(2 )m

ii i i

i i

t px t a g f t

s

(13)

For time scaling, the time-position and scale factors are modified to give

, i i i ip p s s (14)

The module, frequency and phase are not changed, so that the time-scaled version is

1

0

cos 2m

ii i i i

i i

t px t a g f t

s

(15)

B. Frequency-scaling of PCG signals

Frequency-scaling by the MP method is performed by scaling the frequency (f) of each

time-frequency atom of the signal by using

f f (16)

where is a scaling constant.

Thus, the frequency-scaled signal can be expressed as

1

0

cos 2m

ii i i i

i i

t px t a g f t

s

(17)

C. Joint time-frequency scaling of PCG signals

Sometimes a joint time-frequency scaling is desired for changing both the time and the

frequency properties of a signal. The transformation is the combination of the time scaling

and the frequency-scaling described above. Thus,


364

1

0

cos 2m

ii i i i

i i

t px t a g f t

s

(18)

Array phonocardiography

The feasibility of applying passive listening array technology to the detection and localization

of audible acoustic signals in the human body is discussed. A primary clinical objective is the

low cost, non-invasive early diagnosis of Coronary Artery Disease (CAD). Referred to as

array Phonocardiography, the technique uses an array of vibration sensors placed non-

invasively on the external chest wall. A wave speed dispersion curve permits focusing of a

scanning beam to an array near field focal point that is scanned over a desired volume to form

an intensity image. A model for blood flow induced vibrations of a diseased coronary artery is

used to compare with experimental, in vitro, results using a urethane phantom. A turbulence

source can be located within 1 cm imaging using both a time delay-and-sum Conventional

Beam Former (CBF) and a reduced rank Adaptive Beam Former (rrABF). The performance

of the rrABF is sensitive to the spatial extent of the radiating near field source; however, the

imposition of a spatial gain constraint mitigates the problem [19]. Irregularities using near

field focused beam forming techniques has been suggested as a procedure that would enhance

listening performance and serve as a non-invasive, diagnostic screening device [20-22].

Neural network classification

A novel method for segmentation of Heart Sounds (HSs) into single cardiac cycle

(S1-Systole-S2-Diastole) using homomorphic filtering and K-means clustering is presented

by Gupta et al. [32]. Homomorphic filtering technique resulted in smooth envelope enabling

easy peak detection. Peak conditioning was performed to remove peaks, which do not

correspond to S1 and S2. K-means clustering of the time intervals between peaks was used to

indicate the occurrence of single cardiac cycles and also to point to missed cycles.

Appreciable S1 and S2 amplitudes as compared to murmurs enhanced the performance of this

algorithm. Feature vectors were formed after segmentation by using Daubechies-2 wavelet

detail coefficients at the second decomposition level. These feature vectors were then used as

input to the neural networks. Grow and Learn (GAL) and Multilayer Perceptron-

Backpropagation (MLP-BP) neural networks were used for classification of three different

HSs (Normal, Systolic murmur and Diastolic murmur). It was observed that the classification

performance of GAL was similar to MLP-BP. However, the training and testing times of

GAL were lower as compared to MLPBP. The proposed framework could be a potential

solution for automatic analysis of HSs that may be implemented in real time for classification

of HSs [32].

Segmentation using nonlinear dynamic analysis and high-frequency decomposition

An effective methodology for segmenting the temporal trace of PCG signals (shown in Fig. 7)

is presented by Quiceno et al. [35]. Initially, inter-beat segmentation is carried out using the

standard bipolar Lead-II of the ECG recording for locating the occurrence of S1. Next, the

intra-beat segmentation is achieved by using Recurrence Time Statistics (RTS), which is

sensitive to changes of the reconstructed attractor in a state space derived from non-linear

dynamic analysis. If the segmentation using RTS fails, an alternative segmentation is

proposed using thresholding over the Shannon envelogram extracted from the high-frequency

decomposition. The database of PCG records, which was used, belongs to the National

University of Colombia. Inter-beat segmentation accuracy was 100% over all PCG

recordings. Taking into account 360 PCG beats, where a set of 180 beats were strongly


365

disturbed by different types of cardiac murmurs, intra-beat segmentation yielded an accuracy

result of 97.7% [35].

Fig. 7 Correctly segmented PCG signal with systolic murmur

(figure redrawn from Quiceno et al. [35])

Recurrence time statistics (RTS)

RTS are used in order to detect abrupt changes in the signal dynamics, corresponding to S1

and S2. An arbitrary state aref is chosen on the trajectory whereupon all recurrences within a

hyper sphere of radius (r) are selected, i.e., Br(aref) = a: ||a − aref || ≤ r. Ψ is then defined as

the total amount of states in the set Br, and related to the information dimension via a power

law, which motivates its ability to detect weak signal transitions based on the amplitude,

period, dimension and complexity of the signal. A sliding window is used to partition the

recorded PCG signal into overlapping segments, and Ψ is calculated for each segment.

The r value is a very important parameter in the detection algorithm. If it is chosen too low,

the hyper sphere would be low on data, and if r is chosen too high, the hyper sphere will

contain misleading information from erroneous parts of the reconstructed state space.

In this work, r is adaptive, and it becomes lower if there is not lobe detection corresponding to

S1 and S2 in Ψ(r) [34].

Wavelet decomposition and Shannon energy

These methods are used in the detection of S2 when it is considered that RTS did not give a

good estimation of the boundaries. Basically, three steps are implemented: high pass filtering,

wavelet decomposition and Shannon energy operator applied to the detail coefficients.

From the knowledge of cardiac functionality and genesis of S1 and S2 sounds, it is known

that aortic valves close with relatively large pressure difference across the valve, which

originates the high frequency content in S2 sound [2]. This is the motivation to use the

approximation coefficients to perform the detection of S2. To extract the signal envelope from

the detail coefficients, the Shannon energy operator is applied [33].

2

2

1

1) log(

N

d j j j

n

E d n d n d nN

(19)

where dj are the jth

level detail coefficients of the wavelet transformed heart sound signal, and

N is the number of samples in the selected window. This technique emphasizes the medium

intensity signal components and attenuates the effect of low intensity components [34].

Using the signal envelopes provided by the Shannon energy, sound lobe boundaries are

identified applying decision rules based on thresholds.


366

Cluster analysis and classification

Amit et al. [39] describe a computational analysis framework for identifying distinct

morphologies of heart sounds and classifying them into physiological states. The analysis

framework is based on hierarchical clustering, compact data representation in the feature

space of cluster distances and a classification algorithm. Amit et al. [39] applied the proposed

framework on two heart sound datasets, acquired during controlled alternations of the

physiological conditions, and analyzed the morphological changes induced to S1, and the

ability to predict physiological variables from the morphology of S1. On the first dataset of

12 subjects, acquired while modulating the respiratory pressure, the algorithm achieved an

average accuracy of 827% in classifying the level of breathing resistance, and was able to

estimate the instantaneous breathing pressure with an average error of 196%. A strong

correlation of 0.92 was obtained between the estimated and the actual breathing efforts.

On the second dataset of 11 subjects, acquired during pharmacological stress tests, the

average accuracy in classifying the stress stage was 867%. The effects of the chosen raw

signal representation, distance metrics and classification algorithm on the performance were

studied on both real and simulated data. The results suggest that quantitative heart sound

analysis may provide a new non-invasive technique for continuous cardiac monitoring and

improved detection of mechanical dysfunctions caused by cardiovascular and

cardiopulmonary diseases.

The S1 can be represented in three different forms, i.e.,

1. Time-domain representation: Direct characterization of the signal as a time series of

sampled amplitude values.

2. Frequency-domain representation: Spectral characterization of the signal obtained by

applying fast Fourier transforms (FFT).

3. Time-frequency representation: Joint time-frequency characterization of the signal

obtained by applying one of the following transforms:

Short-time Fourier transform (STFT), defined by:

2, i ftS t f s t e d

(20)

S-transform (ST), defined by [43]:

2 2( ) /2 2,

2π

t f i ftfS t f s t e e d

(21)

Wigner-Ville Distribution (WVD), defined by [44];

* 2,2 2

i ftS t f s t s t e d

(22)

Choi-Williams Distribution (CWD), defined by [44];

2 22 ( ) /4 *

2,

4 2 2

i ft tS t f e e s s d d

(23)


367

Hierarchical clustering was applied to S1 signals, using each of the signal representations

described above. The purpose of clustering is to partition a dataset into disjoint subsets

(clusters), such that data elements within the same cluster share some sort of similarity.

Similarity between data elements is measured using a distance metric that is suitable for the

nature of the analyzed data. Two distance metrics were considered in this study:

1. Euclidean distance:

2 2 ( )ST t t t t

t

D s r s r (24)

where st and rt are signals of length n.

2. Cross-correlation:

2 2

1

( ) ( )

t ttST

t t

t t

s ś r ŕD

s ś r ŕ

(25)

where 1

1 n

t

t

ś sn

, 1

1 n

t

t

ŕ rn

.

Clustering was done using an agglomerative hierarchical clustering procedure that initially

partitions a set of n data elements into n clusters, each containing one data element, and then

iteratively merges the two most similar clusters, until the entire dataset forms a single cluster

[45]. The bottom of the created hierarchical tree can next be pruned so that the required

number of clusters N is obtained. Each cut of the data elements are assigned to a single

cluster, creating the output data partitioning to clusters {C1, …, CN}. The algorithm requires a

cluster similarity criterion for choosing the next two clusters to be merged. Authors in [46]

have used Ward’s step-wise optimal criterion, which chooses the clusters such that the

increase in the overall sum-of-squared error after the merge is minimal. The distance between

clusters Ci and Cj is defined by: , / w i j i j i j i jD C C n n n n m m , where ni, and nj are

the sizes of clusters, and mi, and mj are their means.

LMS based least square SVM classifier

Here, a technique to improve the performance of the Least Square Support Vector Machine

(LSSVM) is proposed for classification of normal and abnormal heart sounds using wavelet

based feature set. In the proposed technique, the Lagrange multiplier is modified based on

Least Mean Square (LMS) algorithm, which in turn modifies the weight vector to reduce the

classification error. The basic idea is to enlarge the separating boundary surface, such that the

separability between the clusters is increased. The updated weight vector is used at the time of

testing. The performance of the proposed systems is evaluated on 64 different recordings of

heart sounds comprising of normal and five different pathological cases. It is found that the

proposed technique classifies the heart sounds with higher recognition accuracy than

competing techniques [59].


368

Hilbert transfer

The key features of PCG are extracted based on the slopes of envelop of Hilbert Transfer after

relocating boundaries with energy envelope segmentation [69].

1 x

H x t dt

(26)

Z x x t H x t (27)

where H(x(t)) is the Hilbert transferred signal. The mode of Z(x) is the original envelope.

The overall accuracy of features extraction is found to be 91.95%. 25 significant clinical

features are introduced, and chosen to make two-kind classification by SVM. In the results of

two-kind classification, the overall accuracy is 91.3%, which is better than 85.23% accuracy

in 100 features of Shannon Energy Envelope. The result shows that features including clinical

signification is of signification for enhancing the accurate rate of PCG classification [70].

Wavelet packet entropy

Wavelet transform is a powerful technique in analyzing nonstationary signals such as PCG

signals [77]. The main advantage of wavelet transform is its varying window size that is

narrow for high frequencies and wide for low frequencies. Therefore, wavelet transform is

much more powerful than the other time frequency analysis techniques such as Discrete

Fourier Transform (DFT) and Short-Term Fourier Transform (STFT), not only for providing

useful time and frequency information, but also for its adaptive time and frequency resolution

[78].

A. Wavelet packet transform (WPT)

WPT is an extension of Discrete Wavelet Transform (DWT) whereby all nodes in the tree

structure are allowed to split further at each level of decomposition. With WPT, both the

approximation and detail coefficients are decomposed into approximation and detail

components. In comparison to DWT, WPT decomposes only the approximation coefficients

of the signal (as shown in Fig. 8). Therefore, features can be generated based on

approximation and detail coefficients at different levels to obtain more information.

The WPT of a signal x(t) is defined as follows:

, /22 2 , 0 2 1n j j j s

p n

R

x x t t p dt n (28)

where n is the channel number, j is the number of decomposition level, or scale parameter,

p is the position parameter, ψn(t) is the mother wavelet, and S is the maximum decomposition

level. After decomposing signal x(t) by WPT, 2S sequences can be produced in the S

th level.

The fast decomposition equation for this kind of WPT is

2 , 1 , 2 n j n j

k p

p Z

x h p k x

2 1, 1 ,2n j n j

k p

p Z

x g p k x

(29)


369

where h(i) and g(i) are wavelet quadrature mirror filter coefficients.

Fig. 8 Wavelet packet tree with corresponding high-pass and low-pass filters

(a = approximation coefficients, d = detail coefficients).

The shaded nodes indicate the node not to be produced by DWT.

Three levels of the wavelet packet decomposition with the high-pass and low-pass filters were

illustrated in Fig. 8. This structure can be continued further to decompose the following

approximations and details to reach to a proper level for representing PCG signals of desired

murmurs. From the literature, it can be concluded that levels 6 to 8 were generally chosen for

analyzing PCG signals of different pathological heart sounds [79-83].

B. Entropy

Different types of entropy such as log, norm, Shannon, sure, and threshold can be used to

characterize the heart sounds. However, for this study the entropy introduced by Vitulano and

Casanova [84] for analyzing 1D signals was utilized. They have transformed the

2D mammographic signal into 1D signal through linear transformation and then applied the

entropy on the 1D signal to generate features for differentiating mammograms with different

pathologies. They did not utilize any signal processing technique to analyze the signal prior to

extract entropy features from the signal. In the current study, the PCG signals were first

analyzed with WPT and then entropy features were generated from the wavelet packet

coefficients. Vitulano and Casanova [84] defined the signal “crest” as the part embraced

between lines parallel to the abscissas axis, in which the ordinates are m and M, m is the

absolute minimum and M is the absolute maximum of the signal. Therefore, the signal crest

included all the points x(t) X(t), so that

m x t M (30)

and crest energy is defined as

, ,M

c i

i m

E x i m M

(31)

Signal entropy can be defined based on Ec as

1 , 0, 1cE ES S

E

(32)


370

where E is signal energy, Ec is crest energy, and S is signal entropy. Signal entropy S is

defined based on one-dimensional signals and it has a potential to be applied on the other

dimensional signals such as PCG signals.

Safara et al. [85] employed wavelet packet transform for heart sound analysis, and the entropy

was calculated for deriving feature vectors. Five types of classification were performed to

evaluate the discriminatory power of the generated features. The best results were achieved by

BayesNet with 96.94% accuracy.

Conclusion The PCG signal confirms, and mostly, refines the auscultation data and provides further

information about the acoustic activity concerning the chronology of the pathological signs in

the cardiac cycle, by locating them with respect to the normal heart sounds. Heart murmurs

are the first signs of cardiac valve disorders. Several studies have been conducted in recent

years to automatically differentiate normal heart sounds from heart sounds with murmurs

using various types of audio features. The heart sound categories would be expanded to

include different murmurs. Few methods are discussed for feasible way to enhance the

accurate rate of Phonocardiogram classification. This review also suggests quantitative heart

sound analysis may provide a new non-invasive technique for continuous cardiac monitoring

and improved detection of mechanical dysfunctions caused by cardiovascular and

cardiopulmonary diseases. The first part of this review represents the significance of PCG

signal and the second part demonstrates the different methodology involved for detection,

classification and analysis of PCG signal.

References 1. Goldberger A. L. (1992). Fractal Mechanisms in the Electrophysiology of the Heart,

IEEE Eng Med Biol Mag, 26, 47-52.

2. Leatham A. (1975). Auscultation of the Heart and Phonocardiography, Churchill

Livingstone, New York.

3. Vepa J. (2009). Classification of Heart Murmurs Using Cepstral Features and Support

Vector Machines, In Engineering in Medicine and Biology Society, Annual International

Conference of the IEEE, 2539-2542.

4. Zhang Y-T., G. Chan, X-Y. Zhang, L. Yip (2006). Heart Sounds and Stethoscopes,

In: Wiley Encyclopedia of Biomedical Engineering, Akay M. (Ed.), Wiley-Interscience,

Wiley and Sons, Hoboken, N.J., 1824-1834.

5. Tilkian A. G., M. B. Conover (2001). Understanding Heart Sounds and Murmurs: with an

Introduction to Lung Sounds, Saunders, Philadelphia, 4th Edition.

6. Kingsley B. (1972). Acoustic Evaluation of Prosthetic Cardiac Valve in the Audio

Spectrum, Journal of Audio Engineering Society, 20(9), 750-755.

7. Geckeler G. D., W. Likoff, D. Mason, R. R. Riesz, C. H. Wirth (1954).

Cardiospectrograms: A Preliminary Report, Amer Heart J, 48, 189-196.

8. McKusic V. A., S. A. Talbot, G. N. Weeb (1954). Spectral Phonocardiography: Problems

and Prospects in the Application of the Bell Sound Spectrograph to Phonocardiography,

Bull Johns Hopkins Hosp, 94, 187-198.

9. Hylen J. C., F. E. Kloster, R. H. Herr, A. Starr, H. E. Griswold (1969). Sound

Spectrographic Diagnosis of Ball Variance, Circulation, XXXIX, 849-858.

10. Aigner V. A., G. Muller, E. Knapp, E. Raas (1973). Schallspektrographische

Untersuchungen an Starr-Edwards-Aortenkl appenprotesen, Z Cardiol, 62, 1105-1113.

(in German)


371

11. Aigner V. A. (1977). Kontursonagraphische Analysen von Herzschall-Phgnomenen und

Tbnen verschiedener Klappenprothesen Fortschr, Med, 95, 2162-2166. (in German)

12. Kagawa Y., S. Nitta, S. Naoshi, S. Komei, Y. Shibota, T. Horiuchi, T. Motonao (1977).

Sound Spectroanalytic Diagnosis of Malfunctioning Prosthetic Heart Valve, Tohoku J

Exp Med, 123, 77-89.

13. Cooley J. W., J. W. Tukey (1965). An Algorithm for Machine Calculation of Complex

Fourier Series, Math Computation, 19, 297-301.

14. Rabiner L. R., R. W Schafer (1978). Digital Processing of Speech Signals, Chapter 6,

Prentice-Hall, Englewood Cliffs, N.J., 250-354.

15. Durand L.-G., J. de Guise, R. Guardo (1980). FFT Techniques for the Spectral Analysis

of Prosthetic Heart Valve Sounds, Proc Annu Symp Comput Appl Med Care, 2,

1105-1111.

16. Nagel J. (1986). New Diagnostic and Technical Aspects of Fetal Phonocardiography,

European Journal of Obstetrics & Gynecology and Reproductive Biology, 23(5-6),

295-303.

17. Mallat S. G., Z. Zhang (1993). Matching Pursuits with Time-frequency Dictionaries,

IEEE Trans Signal Processing, 41, 3397-3415.

18. Zhang X., L.-G. Durand, L. Senhadji, H. C. Lee, J.-L. Coatrieux (1998). Time-frequency

Scaling Transformation of the Phonocardiogram Based of the Matching Pursuit Method,

IEEE Transactions on Biomedical Engineering, 45(8), 972-979.

19. Owsley N. L. (2000). Array Phonocardiography, The IEEE 2000 Adaptive Systems for

Signal Processing, Communications, and Control Symposium, AS-SPCC, 31-36.

20. Austin A., A. Hull, N. Owsley, M. Peloquin (1997). Device and Method for Locating

Flow Blockage in a Three-dimensional Object, United States Patent 5,617,869,

Filed: 16 June 1995, Issued: 8.

21. Owsley N. L. (1998). Method and Apparatus for Noninvasive Detection and Analysis of

Turbulent Flow in a Patient's Blood Vessels, United States Patent 5,727,561,

Filed: 23 April 1996, Issued: I7.

22. Qwsley N. L., A. Hull (1998). Beam Formed Near Field Imaging of a Simulated

Coronary Artery Containing a Stenosis, IEEE Trans on Medical Imaging, 37(6), 900-909.

23. Rice M. L., D. J. Doyle (1995). Comparison of Phonocardiographic Monitoring Location,

IEEE 17th Annual Conference Engineering in Medicine and Biology Society, 1, 685-686.

24. Hansen P. B., A. A. Luisada, D. J. Miletich (1989). Phonocardiography as a Monitor of

Cardiac Performance during Anesthesia, Anesth Anulg, 68, 385-387.

25. Durand L. G., Y. E. Langlois, T. Lanthier (1990). Spectral Analysis and Acoustic

Transmission of Mitral and Aortic Valve Closure Sounds in Dogs, Part 4: Effects of

Modulating Cardiac Inotropy, Med Biol Eng Comput, 28, 439-445.

26. Luisada A. A., A. Singhal, F. Portaluppi, C. Strozzi (1985). Noninvasive Index of Cardiac

Contractility during Stress Testing: A Collaborative Study, Clin Cardiol, 8(7), 375-384.

27. Luisada A. A., A. Singhal, V. Knigten (1986). New Index of Cardiac Contractility during

Stress Testing with Treadmill, Acta Curdiol, 41(1), 31-39.

28. Luisada A. A., A. Singhal, F. Portaluppi (1985). Assessment of Left Ventricular Function

by Noninvasive Methods, Adv Curdiol, 32, 111-141.

29. Xiao S., Sh. Cai, G. Liu (2000). Studying the Significance of Cardiac Contractility

Variability Analyzing the Influence of Respiration, Exercise, and Psychological Activity

Using Heart-sound Analysis, IEEE Engineering in Medicine and Biology Magazine,

19(3), 102-105.

30. Jain A. K., A. Ross, S. Prabhakar (2004). An Introduction to Biometric Recognition,

IEEE Trans Circuits Syst Video Technol, Special Issue Image- and Video-based Biomet,

14(1), 4-20.

http://ieeexplore.ieee.org/xpl/mostRecentIssue.jsp?punumber=7076



372

31. Beritelli F., S. Serrano (2007). Biometric Identification Based on Frequency Analysis of

Cardiac Sounds, IEEE Transactions on Information Forensics and Security, 2(3),

596-604.

32. Gupta C. N., R. Palaniappan, S. Swaminathan, S. M. Krishnan (2007). Neural Network

Classification of Homomorphic Segmented Heart Sounds, Applied Soft Computing, 7,

286-297.

33. El-Segaier M., O. Lilja, S. Lukkarinen, L. Srnmo, R. Sepponen, E. Pesonen (2005).

Computer-based Detection and Analysis of Heart Sound and Murmur, Ann Biomed Eng,

33(7), 937-942.

34. Ahlstrom C., O. Liljefeldt, P. Hult, P. Ask (2005). Heart Sound Cancellation from Lung

Sound Recordings Using Recurrence Time Statistics and Nonlinear Prediction, IEEE

Signal Processing Letters, 12(12), 812-815.

35. Quiceno A. F., E. Delgado, M. Vallverd, A. M. Matijasevic, G. Castellanos-Domnguez

(2008). Effective Phonocardiogram Segmentation Using Nonlinear Dynamic Analysis

and High-frequency Decomposition, Computers in Cardiology, 35, 161-164.

36. Tavel M. E. (1978). Clinical Phonocardiography & External Pulse Recording, Third ed.,

Year Book Medical Publishers Inc., Chicago.

37. Sakamoto T. (1965). Hemodynamic Determinants of the Amplitude of the First Heart

Sound, Circ Res, 16, 45-57.

38. Clarke W. B., S. M. Austin, M. S. Pravib, P. M. Griffen, J. T. Dove, J. McCullough,

B. F. Schreiner (1978). Spectral Energy of the First Heart Sound in Acute Myocardial

Ischemia, Circulation, 57, 593-598.

39. Amit G., N. Gavriely, N. Intrator (2009). Cluster Analysis and Classification of Heart

Sounds, Biomedical Signal Processing and Control, 4, 26-36.

40. Durand L. G., M. Blanchard, G. Cloutier, H. N. Sabbah (1990). Comparison of Pattern

Recognition Methods for Computer Assisted Classification of Spectra of Heart Sounds in

Patients with a Porcine Bioprosthetic Valve Implanted in the Mitral Position, IEEE Trans

Biomed Eng, 37(12), 1121-1129.

41. Guo Z., L. G. Durand, H. C. Lee, L. Allard, M. C. Grenier, P. D. Stein (1994). Artificial

Neural Networks in Computer-assisted Classification of Heart Sounds in Patients with

Porcine Bioprosthetic Valves, Med Biol Eng Comput, 32(3), 311-316.

42. Bentley P. M., P. M. Grant, J. T. McDonnell (1998). Time-frequency and Time-scale

Techniques for the Classification of Native and Bioprosthetic Heart Valve Sounds, IEEE

Trans Biomed Eng, 45(1), 125-128.

43. Stockwell R. G., L. Mansinha, R. P. Lowe (1996). Localization of the Complex

Spectrum: The S-transform, IEEE Trans Signal Process, 44(4), 998-1001.

44. Cohen L. (1989). Time-frequency Distributions – A review, Proceedings of the IEEE,

77(7), 941-981.

45. Johnson S. C. (1967). Hierarchical Clustering Schemes, Psychometrika, 2, 241-254.

46. Ward J. H. (1963). Hierarchical Grouping to Optimize an Objective Function, J Am Stat

Assoc, 58(301), 236-244.

47. Hanna I. R., M. E. Silverman (2002). A History of Cardiac Auscultation and Some of its

Contributors, The American Journal of Cardiology, 90, 259-267.

48. Bender J. R. (1992). Yale University School of Medicine Heart Book, New York:

William Morrow and Company, Inc., 167-175.

49. Tavel M. E. (1996). Cardiac Auscultation: A Glorious Past-but does it Have a Future?

Circulation, 93, 1250-1253.

50. Shub C. (2003). Echocardiography or Auscultation? How to Evaluate Systolic Murmurs?

Canadian Family Physician, 49, 163-167.


373

51. Chizner M. A. (2008). Cardiac Auscultation: Rediscovering the Lost Art, Current

Problems in Cardiology, 33, 326-408.

52. Cromwell L., F. J. Weibell, E. A. Pfeiffer (2002). Biomedical Instrumentation and

Measurements, 2nd Ed., PHI Publication, 169-172.

53. Rangayyan R. M. (2002). Biomedical Signal Analysis, IEEE Press, John Wiley & Sons

Inc, 34-38.

54. Cathers I. (1995). Neural Network Assisted Cardiac Auscultation, Artificial Intelligence

in Medicine, 7, 53-66.

55. Ölmez T., Z. Dokur (2003). Classification of Heart Sounds Using an Artificial Neural

Network, Pattern Recognition Letters, 24, 617-629.

56. Reed T. R., N. E. Reed, P. Fritzson (2004). Heart Sound Analysis for Symptom Detection

and Computer-aided Diagnosis, Simulation Modeling Practice and Theory, 12, 129-146.

57. Ari S., G. Saha (2007). On a Robust Algorithm for Heart Sound Segmentation, Journal of

Mechanics in Medicine and Biology, 7(2), 129-150.

58. Choi S. (2008). Detection of Valvular Heart Disorders Using Wavelet Packet

Decomposition and Support Vector Machine, Expert Systems with Application, 35,

1679-1687.

59. Ari S., K. Hembram, G. Saha (2010) Detection of Cardiac Abnormality from PCG Signal

Using LMS Based Least Square SVM Classifier, Expert Systems with Applications, 37,

8019-8026.

60. Xu J., L. Durand, P. Pibarot (2000). Nonlinear Transient Chirp Signal Modeling of the

Aortic and Pulmonary Components of the Second Heart Sound, IEEE Trans Biomed Eng,

47(7), 1328-1335.

61. Xu J., L. Durand, P. Pibarot (2001). Extraction of the Aortic and Pulmonary Components

of the Second Heart Sound Using a Nonlinear Transient Chirp Signal Model, IEEE Trans

Biomed Eng, 48(3), 277-283.

62. Baykal A., Y. Z. Ider, H. Koymen (1995) Distribution of Aortic Mechanical Prosthetic

Valve Closure Sound Model Parameters on the Surface of the Chest, IEEE Trans Biomed

Eng, 42(4), 358-370.

63. Koymen H., B. K. Altay, Y. Z. Ider (1987). A Study of Prosthetic Heart Valve Sounds,

IEEE Trans Biomed Eng, 34(11), 853-863.

64. Sava H. P., J. T. E. McDonnell (1995). Modified Forward-backward Over Determined

Prony’s Method and its Application in Modeling Heart Sounds, IEEE Proceedings –

Vision, Image and Signal Processing, 142(6), 375-380.

65. Leung T. S., P. R. White, W. B. Cook (1998). Analysis of the Second Heart Sound for

Diagnosis of Paediatric Heart Disease, IEEE Proc. – Sci Meas Technol, 145(6), 285-290.

66. Maneejiraprakarn N., P. Panpho, D. Boonthong, P. Thitiwuthikiat, J. Koseeyaporn,

P. Wardkein (2006). Low Cost Non-invasive Instrument for Heart Disease and Peripheral

Vascular Disease Detection, Biomedical Circuits and Systems Conference, IEEE

Biomedical Circuits and Systems Conference, 49-52.

67. Debbal S. M., F. Bereksi-Reguig (2008). Computerized Heart Sounds Analysis,

Computers in Biology and Medicine, 38, 263-280.

68. Debbal S. M., F. Bereksi-Reguig (2007). Automatic Measure of the Split in the Second

Cardiac Sound by Using the Wavelet Transform Technique, Computers in Biology and

Medicine, 37, 269-276.

69. Tu Z., G. Cao, Q. Li, X. Zhang, J. Shi (2010). Improved Methods for Detecting Main

Components of Heart Sounds, Sixth International Conference on Natural Computation, 7,

3585-3588.


http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.Qiao%20Li.QT.&newsearch=true

http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.Xianxia%20Zhang.QT.&newsearch=true

http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.Jun%20Shi.QT.&newsearch=true


374

70. Hu X.-J., G.-T. Cao, H.-H. Zhu, H. Li (2011). Feature Extraction and Choice in PCG

Based on Hilbert Transfer, 4th International Congress on Image and Signal Processing,

IEEE, 4, 2159-2163.

71. Rangayyan R. M., R. J. Lehner (1988). Phonocardiogram Signal Analysis: A Review,

CRC Critical Reviews in Biomedical Engineering, 15(3), 211-236.

72. Durand L. G., M. Blanchard, G. Cloutier, H. N. Sabbah, P. D. Stein (1990). Comparison

of Pattern Recognition Methods for Computer-assisted Classification of Spectra of Heart

Sounds in Patients with a Porcine Bioprosthetic Valve Implanted in the Mitral Position,

IEEE Trans Biomed Eng, 37(12), 1121-1129.

73. Durand L. G., M. Blanchard, H. N. Sabbah, P. D. Stein (1991). Comparison of Spectral

Techniques for Computer Assisted Classification of Spectra of Heart Sounds in Patients

with a Porcine Bioprosthetic Valves, Med Biol Eng Comput, 31(3), 229-236.

74. Durand L. G., P. Pibarot (1995). Digital Signal Processing of the Honocardiogram:

Review of the Most Recent Advancements, Critical Reviews in Biomedical Engineering,

23(3/4), 163-219.

75. Sarbandi R. R., D. D. Doyle, M. Navidbakhsh, K. Hassani, H. Torabiyan (2011). Color

Spectrographic Phonocardiography (CSP) Applied to the Detection and Characterization

of Heart Murmurs: Preliminary Results, BioMedical Engineering OnLine, 10:42,

doi: 10.1186/1475-925X-10-42.

76. Akbari M. A., K. Hassani, J. D. Doyle, M. Navidbakhsh, M. Sangargir, K. Bajelani,

Z. S. Ahmadi (2011). Digital Subtraction Phonocardiography (DSP) Applied to the

Detection and Characterization of Heart Murmurs, BioMedical Engineering OnLine,

10:109, doi: 10.1186/1475-925X-10-109.

77. Ergen B., Y. Tatar, H. O. Gulcur (2011). Time-frequency Analysis of Phonocardiogram

Signals Using Wavelet Transform: A Comparative Study, Computer Methods in

Biomechanics and Biomedical Engineering, 15, 371-381.

78. Chen Y., S. Wang, C.-H. Shen, F. K. Choy (2012). Matrix Decomposition Based Feature

Extraction for Murmur Classification, Medical Engineering & Physics, 34(6), 756-761.

79. Choi S., Y. Shin, H. K. Park (2011). Selection of Wavelet Packet Measures for

Insufficiency Murmur Identification, Expert Systems with Applications, 38(4),

4264-4271.

80. Ahlstrom C., P. Hult, P. Rask (2006). Feature Extraction for Systolic Heart Murmur

Classification, Annals of Biomedical Engineering, 34(11), 1666-1677.

81. Choi S. (2008). Detection of Valvular Heart Disorders Using Wavelet Packet

Decomposition and Support Vector Machine, Expert Systems with Applications, 35(4),

1679-1687.

82. Debbal S. M., F. Bereksi-Reguig (2007). Time-frequency Analysis of the First and the

Second Heartbeat Sounds, Applied Mathematics and Computation, 184(2), 1041-1052.

83. Homaeinezhad M. R., S. A. Atyabi, E. Deneshvar, A. Ghaffari, M. Tahmasebi (2011).

Optimal Delineation of PCG Sounds via False-alarm Bounded Segmentation of a

Wavelet-based Principal Components Analyzed Metric, International Journal for

Numerical Methods in Biomedical Engineering, 27, 1711-1739.

84. Vitulano S., A. Casanova (2008). The Role of Entropy: Mammogram Analysis, Image

Analysis and Recognition, 5112, 863-872.

85. Safara F., S. Doraisamy, A. Azman, A. Jantan, S. Ranga (2012). Wavelet Packet Entropy

for Heart Murmurs Classification, Advances in Bioinformatics, Volume 2012, Article ID

327269.

86. Moukadem A., A. Dieterlen, C. Brandt (2012). Phonocardiogram Signal Processing

Module for Auto-diagnosis and Telemedicine Applications, eHealth and Remote

Monitoring, El Hassani A. H. (Ed.), InTech, doi: 10.5772/48447.

http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.Cao%20Gui-Tao.QT.&newsearch=true


375

87. Sa-Ngasoongsong A., J. Kunthong, V. Sarangan, X. Cai, S. T. S. Bukkapatnam (2012).

A Low-cost, Portable, High-throughput Wireless Sensor System for Phonocardiography

Applications Sensors, 12, 10851-10870, doi: 10.3390/s120810851.

88. Hult P., T. Fjällbrant, B. Wranne, P. Ask (2004). Detection of the Third Heart Sound

Using a Tailored Wavelet Approach, Med Biol Eng Comput, 42(2), 253-258.

89. Wiggers C. J. (1919). Factors Determining the Relative Intensity of the Heart Sounds in

Different Auscultation Areasan Experimental Study, Arch Intern Med, 24(5), 471-488.

90. Dock W. (1933). Mode of Production of the First Heart Sound, Arch Intern Med, 51(5),

737-746.

91. Dean A. L (1916). The Movements of the Mitral Cusps in Relation to the Cardiac Cycle,

Amer J Physiol, 40, 206-217.

92. Wolfenh C. C., A. Margolies (1945). Heart Sounds, In: Diagnosis and Treatment of

Cardiovascular Disease, Stroud W. D. (Ed.), 3rd Ed., Philadelphia, F. A. Davis.

Assist. Prof. Akash Kumar Bhoi, M.Tech


Akash Kumar Bhoi has completed his B.Tech. (Biomedical

Engineering) from the TAT, Bhubaneswar and M.Tech. (Biomedical

Instrumentation) from Karunya University, Coimbatore in the year

2009 and 2011 respectively. He is pursuing Ph.D. from Sikkim

Manipal University, India. He is working as an Assistant Professor in

the Department of Applied Electronics and Instrumentation at Sikkim

Manipal Institute of Technology (SMIT), India since 2012. He is a

member of ISEIS and IAENG, an Associate member of UACEE and

Editorial Board member of IJAEEE, IJAEEE, ITSI-TEEE, IJECCE

and IJCTT and a reviewer of IJET, IJCT, IJBSE, IJMI, AJBLS.

His areas of research are biomedical signal processing, medical

image processing, sensor and transducer and medical

instrumentation. He has published papers in national and

international journals and conferences. He has also published 3 book

chapters. He has participated in workshops, seminars and

conferences.



376

Prof. Karma Sonam Sherpa, Ph.D.


Karma Sonam Sherpa has completed his B.E. (Electrical

Engineering) from the MREC, Jaipur, and M.Tech. (Power

Electronics and Machine Drives) from IIT, Kharagpur in the year

1996 and 2003 respectively. He has a Ph.D. from Sikkim Manipal

University. He has been serving SMIT, Sikkim for the last sixteen

years. Presently, he is a Professor in the Department of Electrical

and Electronics Engineering and officiating head of the

Department of Electrical and Electronics Engineering and

Department of Applied Electronics and Instrumentation

Engineering. He is a life member of ISTE, IEI and System Society

of India. His areas of interests are electric power distribution

system, power electronics and electrical drives. He has published

papers in national and international journals and conferences.

He has organized workshops, seminars and conferences.

Prof. Dr. Bidita Khandelwal, MD.


Dr. Bidita Khandelwal did her MD Medicine in 1998 from Assam

University. She joined Sikkim Manipal Institute of Medical Sciences

in 2000 as an Assistant Professor in Department of Medicine and

since 2009 heads the Department. Since 2015 Dr. Bidita Khandelwal

is a Professor. She is a life member of API, IMA, IACM and RSSDI.

Other posts held are Chairperson of complaint committee on Sexual

Harassment at work place, Chairperson of IEC, Ayurveda Regional

Research Institute, Vice chairman API East Zone, Editorial Board

member of IJEM, a Member of Academic Senate of SMU, a Member

of Sikkim Medical Council, Coordinator post graduate studies and an

External expert for Board of Study. She is a Member of Research

Protocol Evaluation Committee, Publication Evaluation committee,

Institutional Ethics Committee, Hospital Management Committee,

Infection Control Committee, Promotion Committee, State Task

Force, Pharmacovigilance Committee. She has special interest in

diabetes mellitus and has several publications related to it.



Documents

Multidimensional Analytical Study of Heart Sounds: A Review